Reverse Diffusion on 2D Swiss Roll Data

Research Report

Github Repo

Project Overview

Diffusion models have gained popularity due to their effectiveness in various creative and practical applications. They have been extensively used in text-to-image generation, layout-to-image generation, image inpainting, and resolution enhancement. Our project simplifies the application of diffusion models to a 2D dataset, specifically the Swiss roll dataset, aiming to clarify the mathematical and visual concepts behind how a basic diffusion model operates.

Forward and Reverse Diffusion Processes for Swiss Roll Data

For full results, please view the Research Report.